Transformer-Based Deep Neural Language Modeling for Construct-Specific Automatic Item Generation

Algorithmic automatic item generation can be used to obtain large quantities of cognitive items in the domains of knowledge and aptitude testing. However, conventional item models used by template-based automatic item generation techniques are not ideal for the creation of items for non-cognitive constructs. Progress in this area has been made recently by employing long short-term memory recurrent neural networks to produce word sequences that syntactically resemble items typically found in personality questionnaires. To date, such items have been produced unconditionally, without the possibility of selectively targeting personality domains. In this article, we offer a brief synopsis on past developments in natural language processing and explain why the automatic generation of construct-specific items has become attainable only due to recent technological progress. We propose that pre-trained causal transformer models can be fine-tuned to achieve this task using implicit parameterization in conjunction with conditional generation. We demonstrate this method in a tutorial-like fashion and finally compare aspects of validity in human- and machine-authored items using empirical data. Our study finds that approximately two-thirds of the automatically generated items show good psychometric properties (factor loadings above .40) and that one-third even have properties equivalent to established and highly curated human-authored items. Our work thus demonstrates the practical use of deep neural networks for non-cognitive automatic item generation.

Transformer-Based Deep Neural Language Modeling for Construct-Specific Automatic Item Generation

Algorithmic automatic item generation can be used to obtain large quantities of cognitive items in the domains of knowledge and aptitude testing. However, conventional item models used by template-based automatic item generation techniques are not ideal for the creation of items for non-cognitive constructs. Progress in this area has been made recently by employing long short-term memory recurrent neural networks to produce word sequences that syntactically resemble items typically found in personality questionnaires. To date, such items have been produced unconditionally, without the possibility of selectively targeting personality domains. In this article, we offer a brief synopsis on past developments in natural language processing and explain why the automatic generation of construct-specific items has become attainable only due to recent technological progress. We propose that pre-trained causal transformer models can be fine-tuned to achieve this task using implicit parameterization in conjunction with conditional generation. We demonstrate this method in a tutorial-like fashion and finally compare aspects of validity in human- and machine-authored items using empirical data. Our study finds that approximately two-thirds of the automatically generated items show good psychometric properties (factor loadings above .40) and that one-third even have properties equivalent to established and highly curated human-authored items. Our work thus demonstrates the practical use of deep neural networks for non-cognitive automatic item generation.

Automating the Generation of Test Items

Increasing demand for knowledge of our workers has prompted the increase in assessments and providing feedback to facilitate their learning. This and the increasingly computerized assessments require new test items beyond the ability for content specialists to produce them in a feasible fashion. Automatic item generation is a promising method that has begun to demonstrate utility in its application. The purpose of this chapter is to describe how AIG can be used to generate test items using the selected-response (i.e., multiple-choice) format. To ensure our description is both concrete and practical, we illustrate template-based item generation using an example from the complex problem-solving domain of the medical health sciences. The chapter is concluded with a description of the two directions for future research.

Notes From the Field: Automatic Item Generation, Standard Setting, and Learner Performance in Mastery Multiple-Choice Tests

Mastery learning assessments have been described in simulation-based educational interventions; however, studies applying mastery learning to multiple-choice tests (MCTs) are lacking. This study investigates an approach to item generation and standard setting for mastery learning MCTs and evaluates the consistency of learner performance across sequential tests. Item models, variables for question stems, and mastery standards were established using a consensus process. Two test forms were created using item models. Tests were administered at two training programs. The primary outcome, the test–retest consistency of pass–fail decisions across versions of the test, was 94% (κ = .54). Decision-consistency classification was .85. Item-level consistency was 90% (κ = .77, SE = .03). These findings support the use of automatic item generation to create mastery MCTs which produce consistent pass–fail decisions. This technique broadens the range of assessment methods available to educators that require serial MCT testing, including mastery learning curricula.

An efficient and adaptive test of auditory mental imagery

The ability to silently hear music in the mind has been argued to be fundamental to musicality. Objective measurements of this subjective imagery experience are needed if this link between imagery ability and musicality is to be investigated. However, previous tests of musical imagery either rely on self-report, rely on melodic memory, or do not cater in range of abilities. The Pitch Imagery Arrow Task (PIAT) was designed to address these shortcomings; however, it is impractically long. In this paper, we shorten the PIAT using adaptive testing and automatic item generation. We interrogate the cognitive processes underlying the PIAT through item response modelling. The result is an efficient online test of auditory mental imagery ability (adaptive Pitch Imagery Arrow Task: aPIAT) that takes 8 min to complete, is adaptive to participant’s individual ability, and so can be used to test participants with a range of musical backgrounds. Performance on the aPIAT showed positive moderate-to-strong correlations with measures of non-musical and musical working memory, self-reported musical training, and general musical sophistication. Ability on the task was best predicted by the ability to maintain and manipulate tones in mental imagery, as well as to resist perceptual biases that can lead to incorrect responses. As such, the aPIAT is the ideal tool in which to investigate the relationship between pitch imagery ability and musicality.